HOI-Diff: Text-Driven Synthesis of 3D Human-Object Interactions using Diffusion Models